Re: Requirements specification

[julian_satran@il.ibm.com]

I did walk (more than once) into a complex network - but instead of looking
at screens
I took a look at logs - and they tell a different story than yours.
A link outage at the middle of a network may result (and usually does) in a
TCP connection failing.
Sometimes this is handled by applications but not always. And yes - the
network does not fail
but this a statistical argument - some applications may loose hours of work
as SCSI was designed with a very reliable interconnect in mind.
A link outage at the end nodes always results in TCP connections getting
dropped.

And failover is handled badly at the SCSI layer (and FCP-2 goes a long way
to avoid percolating errors to SCSI).

Julo

David Robinson <David.Robinson@EBay.Sun.COM> on 09/08/2000 22:11:00

Please respond to David Robinson <David.Robinson@EBay.Sun.COM>

To:   ips@ece.cmu.edu
cc:    (bcc: Julian Satran/Haifa/IBM)
Subject:  Re: Requirements specification




> It would be nice if you would share with us:
>
>    what are the "wonderful rerouting properties" you are referring to ?
>    what is the link technology to do failover ?
>
> I am sure only about myself but I guess that many of us are ignorant on
the
> above.

Walk up to any well designed IP network (some ISP's machine room) and
randomly cut a network cable, everything just keeps working.  Routers
detect dead links and use another redundant link.  IP technology is
giving the Pentagon headaches, no longer does bombing one facility
wipe out a command and control network. Cheap technology from Cisco
defeats million dollar bombs!

> As for availability - several connections and a failover mechanism will
> make
> a solution more reliable. The question is who is doing the failover.

Doing failover between multiple parallel or FC SCSI channels is very
well known and in wide practice today.  Why is it then a requirement
that we put it into iSCSI?

     -David